Feature Selection and Recognition of Multivariate Physiological Signals Using Binary Firefly Algorithm

• Published in: IEEE International Conference on Industrial Informatics (INDIN) Vol. 1; pp. 821 - 826
• Main Authors: He, Hong, Peng, Feihu, Ying, Jun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2019
• Subjects: Accuracy; Blood pressure; classification; Classification algorithms; ECG; Electrocardiography; Feature extraction; feature selection; firefly algorithm; Heart beat; multiple physiological signal; Physiology; Pressure measurement; Principal component analysis; Telemedicine
• ISSN: 2378-363X
• DOI: 10.1109/INDIN41052.2019.8972111

Interest in pattern recognition of physiological signals is growing in recent years due to the rapid development of telemedicine and the Internet of Things. However, on account of high dimensionality, multivariate, and the large amount of data, feature extraction and selection of multiple physiological signals is still a hard nut to crack in pattern recognition. Therefore, a measuring experiment of multivariate physiological signals before and after physical exercises was firstly carried out. Signals of electrocardiogram (ECG), blood pressure and oxygen saturation are simultaneously measured for 30 subjects. Moreover, multivariate physiological signals are in turn processed by filtering, segmentation and morphological feature extraction based on heartbeat samples. A novel binary firefly algorithm (NBFA) is proposed in this paper to realize the automatic feature selection for multiple physiological signals. The NBFA evaluates each binary firefly according to both classification accuracy and the number of features. The movement of a firefly is updated by a bit conversion factor and a random mounting flying step. High classification results of 30 subjects show that morphological features can depict main characteristics of 12-lead ECG signals. The NBFA performs superior to the PCA, PSO in the feature selection of multiple physiological signals. It can quickly find the small number of features as well as obtain high classification, which further reduces the computation load and data storage capacity in pattern recognition of multiple physiological signals.